Face Recognition Research Based On Histograms Of Nonsubsampled Contourlet Oriented Gradient

Face recognition technology has now become a very hot research topic for its potential applications.As an important branch of biometrics, face recognition has a very large value in national security, military security, public safety management, smart guard, immigration control, authentication, and human-computer interaction field,because of its non-contact nature, universality, easy collection,Existing face recognition systems can be achieved satisfactory results with user cooperation, acquisition conditions ideal situation, but when the user does not fit and the acquisition conditions are not ideal, the recognition performance will drop sharply, therefore, there are still a number of key issues need to be addressed to take people face recognition technology into practice.A face recognition system generally consists of face detection, facial feature extraction, feature classificatio. Facial feature extraction is the most critical step of the face recognition system, because the feature extraction from face image determines the final recognition rates radically in this step. Non-subsampled Contourlet transform can decomposed face image into several non-sampling Contourlet coefficients, including high-frequency coefficients and low frequency coefficients, low frequency coefficient contains the basic characteristics of the human face images, and high-frequency coefficients contain the human face texture and orientation information, and the high-frequency coefficients is not particularly sensitive to changes in illumination, illumination problem can be solved to some extent, Histogram of Oriented Gradient(HOG) can extract the edge direction features of face images, and because it uses a window of statistics, the algorithm is robustness to gestures, facial expressions, etc.This paper will combine NSCT transform and HOG and applied to face recognition, mainly as follows:(1)Proposed a new face recognition algorithm based on Histogram of NSCT Oriented Gradient(HNOG), we analyze the influence of various parameters and find the optimal parameters of HNOG algorithm through a large number of experiments, Experimental results using multi-channel nearest neighbor classifier based on Euclidean distance show that, the extracted feature using the proposed method is robust to the variation of illumination, expression and pose, and has better recognition performance.(2)Light has been a disaster division of face recognition, HOG though to a certain robustness to illumination changes, but dramatic changes in the light, still can not get enough recognition features, and relatively light gradient can well overcome with to the influence of the traditional relative gradient histogram is improved and alternative HNOG algorithm HOG, presented nonSubsampled contourlet relative gradient direction histogram characterization method(HNROG). Experimental results show that this feature extraction method is characterized very well eliminate the effects of dramatic lighting changes, can effectively improve the recognition rate of the algorithm.